SAN FRANCISCO Intel Corp. has revealed the details of two advanced memory technologies that it plans to position as successors to flash memory. While neither of the concepts is anywhere close to making the transition from a research idea to a commercial product, some analysts say that if the parts perform as advertised, they could be exactly what the nonvolatile memory market needs: a low-cost device with reasonably fast read and write times that is easy to manufacture.
"The industry has been searching for a new kind of memory that would be the Holy Grail of memories," said Stefan Lai, vice president of Intel's technology and manufacturing group, and the original developer of the company's flash technology. He noted that all of the existing mainstream memories have some limiting factors. DRAM is cheap, but it is volatile and difficult to integrate with logic components. SRAM is another volatile technology and is expensive. And flash is nonvolatile, but slower than the other types and has a finite number of write cycles.
To solve these problems, Intel's researchers have been looking at several new types of storage devices. One of these is polymeric ferroelectric RAM (PFRAM), or polymer memory, which uses two layers of metal strands running at perpendicular angles with a thin polymer sheet sandwiched in between. "This is basically a layer of plastic," said Lai.
He said that each memory cell is formed at the intersection between the crisscrossing metal, and the data is stored by changing the polarization of the polymer between the metal lines. This design would require only a single substrate of CMOS circuitry at the bottom, and up to eight layers of polymer could be built atop the base. Current memory designs require one transistor to control every memory cell, but with polymer memory there are no transistors at all. Best of all, Lai said the polymers could be manufactured with a simple spin-on process, which is very inexpensive.
"This would cost about one-eighth the cost of CMOS memory," he said. "It is very inexpensive to make." With a write time of about 50 microseconds, it is slower than the single-microsecond write time of NAND flash, but it is still fast enough for most consumer applications, and Lai said the price difference would make it a preferred choice of current flash technology. "We think this is a very good technology for removable memory cards," he said. Intel has not disclosed any data on when these new memories may become commercially available.
The second type of memory that Intel is developing is termed Ovonyx unified memory (OUM), after the company, Ovonyx Inc. that owns much of the intellectual property in the field and is working with Intel to fully develop the concept. OUM is also built atop a CMOS silicon wafer, but then it uses a thin film of a special material called chalcogenide. This is the same material that is currently used in rewriteable CD-ROMs and DVD, according to Tyler Lowrey, president and chief executive of Sunnyvale, Calif.-based Ovonyx.
According to Lai, chalcogenide changes its state when it is subjected to lasers, for CDs and DVD, or to electric current, for the OUM devices. By passing electricity through the material, the atoms within shift from an amorphous state with random positioning to a crystalline state with more ordered positioning. This can change the capacitance of the material by up to 200 percent, which is how the data can be stored in the cells.
To date, the company has made this concept work in single cells and has produced a 4-Mbit test chip. However, Lai stressed that a lot of work remains before it can be commercialized. Even so, the early results show much promise. With read and write times of 100 nanoseconds, it is much faster than current flash technologies, and just as nonvolatile. Lai said it is less expensive to produce than flash and easier to integrate into standard logic processes. "I don't see anything that says this won't work," he said. "We are ready to see if we can make this into a high-volume product."